An energy benchmarking model based on artificial neural network method utilizing US Commercial Buildings Energy Consumption Survey (CBECS) database

This study focuses on development of an energy benchmarking model utilizing U.S. Commercial Buildings Energy Consumption Survey (CBECS) Database. An artificial neural networks (ANN) method based approach was used in the study. Office type buildings in the CBECS database were used in the benchmarking model development and weighted energy use intensity (EUI) was selected as the benchmarking index. The benchmarking model included input variables describing building's physical properties, occupancy and climate. Yearly electricity consumption per square meter, or EUI, was estimated by the ANN model. The correlation coefficient for each census division benchmarking model varied between 0.45 and 0.73, and mean squared error (MSE) varied between 9.60 and 15.25. It was observed that when the data set for a census division was grouped by different climate zones, ANN benchmarking model provided more accurate predictions. It was also observed that ANN model provides more accurate estimations when compared with predictions obtained with multi-linear regression models. For comparison, the MSE values varied between 10.24 and 40.43. Overall, the ANN model proved itself a better prediction model for energy benchmarking. Copyright © 2006 John Wiley & Sons, Ltd.